公众号记得加星标⭐️，第一时间看推送不会错过。 来源：内容 编译自 androidauthority 。 USB-C 的本意是简化一切——但它失败了。 USB-C 亮相已经很久了。从统一的、一刀切的规范，到如今兼容性问题和功能支持不透明的严峻现 实，USB-C 有赞有贬。我坚定地站在中间——意识到问题的存在，但仍然希望——尽管这种希望很 愚蠢——这个值得信赖的端口有一天能兑现它的承诺。不幸的是，随着时间的推移，USB-C 的机会 之窗正在迅速关闭。 要确切了解 USB-C 的"问题"所在，只需环顾一下你的客厅。你能记得哪个移动电源给哪个设备充电 快或慢吗？笔记本电脑和台式电脑也好不到哪里去。以前我们有 DisplayPort、HDMI 和圆形接口的 时候，你很清楚自己的情况——但现在，要弄清楚如今三四个 USB-C 接口哪个能用，就需要仔细阅 读说明书了。谁有时间呢？ 玩"猜猜我是谁？"游戏，用一个号称万能但实际却很少用的接口，只是 USB-C 最大问题的一个缩影 ——其规范本身就很混乱。如果有人能告诉我智能手机领域还有多少种不同的充电标准，或者苹果 Mac 系列产品中有多少种不同的数据传输速度，那真是太好了 ...

公众号记得加星标⭐️，第一时间看推送不会错过。 来源：内容 编译自 techspot 。 在我们对英伟达 RTX 5060 Ti 8GB 版本的评测中，我们曾指出GeForce 50 系列可能会成为该公司 有史以来最差的产品。然而，尽管价格更合理，产品稀缺，AMD 的竞争激烈，以及诸多批评，Team Green 的最新显卡系列仍使其在市场上占据了前所未有的主导地位（即使不考虑人工智能的影响）。 根据 Jon Peddie Research 的数据， 2025 年第一季度，专用桌面显卡的总销量达到 920 万台，同比 增长和环比增长均超过 8%。尽管 Nvidia 和 AMD 在此期间都推出了新的 GPU 产品线，但 Nvidia 的收益要大得多。 英伟达在1月至3月期间售出了数百万台RTX 50系列GPU，而AMD的Radeon 9000系列出货量不足75 万台。英伟达的市场份额飙升至历史性的92%，将AMD挤压至8%的历史低点，英特尔几乎寸步难 行。乔恩·佩迪表示，AMD的产量不足是造成这一巨大差距的主要原因。 AMD 在 3 月份报告称RX 9070 和 9070 XT 的需求"前所未有"，这表明该公司措 ...

Core Viewpoint - Intel is developing new chip packaging technologies to enhance processor capabilities for artificial intelligence, addressing the limitations imposed by Moore's Law and the fixed size of individual silicon chips [1]. Group 1: Innovations in Chip Packaging - Intel announced three innovations aimed at increasing the number and size of silicon chips within a single package, allowing integration of over 10,000 square millimeters of silicon in a package area exceeding 21,000 square millimeters [1]. - The latest EMIB technology, EMIB-T, enhances connection density with vertical copper connections (TSV) and includes a copper mesh for grounding, enabling the connection of 38 or more EMIB-T bridges to over 12 full-size silicon chips [3][4]. Group 2: Thermal Management Solutions - Intel introduced low thermal gradient thermal compression bonding, which improves the predictability and management of thermal expansion mismatches, allowing for the installation of chips on oversized substrates [6]. - The company is addressing the challenge of heat dissipation from larger silicon components by developing modular integrated heat sinks, which can maintain flatness and reliability at higher temperatures [9]. Group 3: Competitive Landscape - These technologies are still in the research and development phase, with no commercial release date announced, but they are expected to be crucial for competing with TSMC's packaging expansion plans in the coming years [9].

Core Viewpoint - TSMC is expanding its advanced packaging capacity, but the timeline for the AP7 plant in Chiayi has been delayed, raising concerns about the impact on global HPC chip supply [1][2]. Group 1: Plant Development and Delays - TSMC's AP7 plant in Chiayi was initially scheduled for equipment installation in Q3, but this has been postponed to Q4 due to recent safety incidents at the site [1][2]. - The Chiayi County Mayor mentioned that the first phase of the plant's construction is expected to proceed in Q3, despite industry reports indicating a delay [1][2]. - The Chiayi plant is set to focus on wafer-level multi-chip module (WMCM) packaging, which is anticipated to be used in Apple's self-developed chips [1][2]. Group 2: Safety Incidents and Their Impact - Two safety incidents involving different contractors have led to work stoppages at the Chiayi plant, which may affect the construction timeline [2]. - The Occupational Safety and Health Administration has stated that TSMC, as the owner, is involved in discussions to enhance safety measures following these incidents [2]. Group 3: Future Production Capacity - TSMC is planning to establish two advanced packaging plants in Chiayi, with recruitment for technical staff already underway, offering salaries above NT$700,000 [2]. - The AP7 plant is expected to have a monthly production capacity of 4,000 to 5,000 units for SoIC by 2024, with potential increases in subsequent years [4]. - TSMC's SoIC technology is set to be utilized by major clients, including AMD and Apple, with the latter expected to implement it in the M5 chip [5]. Group 4: Overall Industry Outlook - Despite concerns about capacity constraints due to TSMC's investments in the U.S., the progress of the AP8 and AP7 plants in Taiwan is reportedly accelerating [4]. - The industry anticipates growth in operations as new advanced packaging facilities come online and testing facilities continue to expand capital expenditures [6].

Core Viewpoint - AheadComputing, a startup founded by former Intel architects, aims to revolutionize microprocessor design using the RISC-V open architecture, moving away from Intel's x86 dominance in the CPU market [1][2][4]. Group 1: Company Overview - AheadComputing was established a year ago by four founders who previously worked at Intel for nearly a century, focusing on developing a new generation of microprocessors [1]. - The company is committed to producing the world's largest and most powerful CPU, challenging Intel's long-standing market dominance [1][5]. - The startup has received $22 million in venture capital funding, indicating strong investor interest in its innovative approach [5]. Group 2: Market Dynamics - The CPU market is becoming fragmented as major tech companies like Apple and Google adopt alternative architectures, contributing to Intel's struggles [1][3]. - RISC-V, the architecture chosen by AheadComputing, is appealing to startups because it is an open technology that does not require licensing fees, unlike x86 or ARM designs [2][3]. - The traditional dominance of large semiconductor companies is declining, allowing new entrants like AheadComputing to carve out niches in the market [3][5]. Group 3: Industry Trends - The semiconductor industry is undergoing a transformation, with a shift towards specialized chips optimized for specific tasks, moving away from the all-encompassing designs of traditional CPUs [5][8]. - The rise of AI and quantum computing is driving innovation in chip design, creating opportunities for companies like AheadComputing to leverage new technologies [5][6]. - The increasing trend of outsourcing manufacturing to foundries like TSMC allows startups to enter the market without the high costs associated with building semiconductor fabrication plants [3][5]. Group 4: Challenges and Opportunities - Despite the potential for high rewards, the transition to new architectures like RISC-V carries significant risks, as the industry is uncertain about which chip designs will prevail [6]. - The departure of experienced engineers from Intel to startups like AheadComputing highlights a shift in talent and innovation within the semiconductor sector [6][9]. - The ongoing challenges faced by Intel, including layoffs and manufacturing setbacks, may open doors for new companies to thrive in the semiconductor landscape [8].

特斯拉的 Dojo 是目前全球最大的两款处理器之一。这些巨大的晶圆级芯片使用整块 300 毫米晶圆， 这意味着一次性构建更大的计算能力根本不可能。每个 Dojo 晶圆级处理器最多可容纳 8,850 个核 心，但其中一些核心在部署后可能会引发静默数据损坏 (SDC)，从而破坏大规模训练运行的结果。 大型处理器 鉴于 Dojo Training Tile（大型晶圆尺寸芯片）的极端复杂性，即使在制造过程中也不容易检测到有 缺陷的芯片，但是当涉及到静默数据损坏（SDC）时，事情会变得更加复杂。 请记住，所有类型的硬件都不可避免地会出现 SDC，但 Dojo 处理器的电流消耗高达 18,000 安培， 功耗高达 15,000 瓦，这会产生影响。不过，所有核心都应该按预期运行，否则特斯拉的 AI 训练将 变得更加复杂，因为数据损坏导致的一个错误就可能使数周的 AI 训练付诸东流。 公众号记得加星标⭐️，第一时间看推送不会错过。 来源：内容 编译自 tesla 。 在大型处理器上检测故障核心并将其禁用是一项挑战，但特斯拉开发了 Stress 工具，该工具不仅可 以在 Dojo 处理器上检测容易出现静默数据损坏的核心，还可以 ...

Core Viewpoint - AMD is heavily investing in acquisitions to enhance its AI capabilities and compete with Nvidia in the data center GPU market, particularly through its Instinct product line [1][2]. Group 1: AMD's AI Strategy and Acquisitions - AMD is focusing on providing "end-to-end AI solutions" and has made several acquisitions over the past three years to bolster its AI strategy, including companies like Mipsology, Nod.ai, Silo AI, and ZT Systems [1][2][3]. - The company plans to unveil its next-generation Instinct GPU, MI400, at the upcoming Advancing AI 2025 event, with positive early customer feedback indicating a significant expansion of its AI accelerator market [1][2]. - AMD's CEO Lisa Su highlighted the importance of these acquisitions in enhancing the company's AI capabilities and addressing the growing demand for AI infrastructure [1][2]. Group 2: Specific Acquisitions and Their Impact - The acquisition of Untether AI aims to improve AMD's AI hardware and software engineering capabilities, focusing on energy-efficient AI inference chips [5][6]. - AMD's purchase of Brium is expected to enhance its AI software capabilities, particularly in optimizing AI solutions for its hardware [8]. - The acquisition of Enosemi will support the development of next-generation AI systems through advanced photonic solutions, improving data transmission efficiency [9][11]. - ZT Systems was acquired for $4.9 billion to integrate its expertise in data center solutions with AMD's chip offerings, aiming to accelerate the deployment of AI systems [10][12]. - The acquisition of Silo AI for $665 million positions AMD to leverage Europe's largest private AI lab for developing open-standard AI solutions [14][15]. - Nod.ai's acquisition is intended to enhance AMD's open software capabilities, facilitating easier deployment of AI solutions optimized for AMD hardware [16][17]. - Mipsology's acquisition will accelerate collaboration with customers and expand AMD's AI software development capabilities [18][19]. - The acquisition of Xilinx for $49 billion is expected to significantly enhance AMD's opportunities in data centers and embedded computing, further driving its AI strategy [22][23].

Core Viewpoint - The article discusses the launch of the Pulsar satellite by Xona Space Systems, which aims to provide a more reliable and precise alternative to the existing GPS system, addressing its vulnerabilities to interference and deception [1][2][10]. Group 1: Satellite Technology and Development - The Pulsar satellite is the first of a planned constellation of 258 satellites designed to operate in low Earth orbit, approximately 12,000 miles above the Earth, offering stronger and more precise signals than GPS [1][2]. - Xona's CTO, Tyler Reid, claims that the signals from their satellites will be about 100 times stronger than GPS signals, allowing for better penetration through buildings and reduced interference coverage [1][2]. - The satellite system aims to achieve positioning accuracy of 3 inches (10 centimeters) or better, significantly improving navigation reliability in challenging environments [3][5][6]. Group 2: GPS Vulnerabilities and Market Need - The GPS system, operational since 1993, is critical for various applications but is susceptible to interference from various sources, including space weather and low-cost jammers [2][8]. - The ongoing conflict in Ukraine has highlighted the vulnerabilities of GPS, as interference has rendered precision-guided munitions ineffective, prompting a search for reliable alternatives [8][10]. - The proliferation of inexpensive jamming devices has raised concerns about GPS reliability, affecting commercial flights and other critical operations [8][10]. Group 3: Competitive Landscape and Future Plans - Xona is not the only company exploring alternatives to GPS; other firms like Anello Photonics and Advanced Navigation are developing ground-based solutions using inertial navigation systems [10][11]. - Xona plans to launch additional satellites in the coming years, with a goal to deploy most of its constellation by 2030, enhancing the resilience of critical infrastructure reliant on GPS [11][12]. - The company aims to design signals compatible with existing GPS technology, facilitating integration for GPS receiver manufacturers [11].

公众号记得加星标⭐️，第一时间看推送不会错过。 在科技浪潮奔涌的当下，全球存储市场早已形成了泾渭分明的竞争版图——三星、SK海力士、美 光 三大巨头宛如鼎立的行业巨擘，牢牢把控着存储行业命脉。 根据TrendForce集邦咨询数据披露，2025年第一季度数据显示，SK海力士以36%的DRAM营收份额 位居榜首，三星（33.7%）与美光（24.3%）紧随其后，三家合计占据全球DRAM市场94%的份额。 在NAND闪存领域，三星以31.9%的市占率领跑，SK海力士（含Solidigm）以16.6%位列第二，美 光、铠侠和闪迪则分别占据15.4%、14.6%和12.9%的市场份额。 英特尔作为半导体领域的老牌霸主，早年曾与美光携手开创闪存时代，却因战略重心向CPU倾斜，在 2025年彻底剥离NAND业务，黯然退出存储主战场。 而日本存储产业的兴衰更堪称行业教科书——上世纪80年代，东芝、日立等企业曾占据全球DRAM市 场半壁江山，却因错失3D堆叠技术先机与成本控制失利，最终在世纪之交的价格战中全线溃败，如 今仅存的铠侠（原东芝存储）也只能在细分市场艰难求生。 存储市场，输家联盟再出发 这些 消逝的行业参与者，恰似存 ...

Core Viewpoint - The article discusses the challenges in semiconductor parameter extraction due to the complexity of device models and the limitations of traditional optimization algorithms, introducing Keysight's ML Optimizer as a revolutionary solution to enhance efficiency and accuracy in this process [1][4]. Group 1: Challenges in Semiconductor Parameter Extraction - The complexity of semiconductor device models has made parameter extraction increasingly challenging, with traditional optimization algorithms often getting stuck in local optima due to unclear gradient changes [1]. - The presence of numerous interrelated parameters in modern semiconductor models leads to inefficiencies in traditional methods, requiring engineers to break down the extraction process into lengthy sub-steps, which can take days or even weeks [1]. Group 2: Introduction of ML Optimizer - Keysight has launched the ML Optimizer, a machine learning-based global optimizer that simplifies the parameter extraction process by synchronously handling vast amounts of data in a single step, reducing the extraction time from days to just hours [2][4]. - The ML Optimizer excels in navigating non-convex parameter spaces, utilizing advanced machine learning algorithms to find global optima more accurately, thereby improving the precision of parameter extraction and the consistency of overall fitting [4]. Group 3: Live Demonstration and Applications - A live demonstration is scheduled to showcase the ML Optimizer's effectiveness in various device modeling tasks, including diodes, GaN HEMTs, MOSFETs, and BJTs, with interactive elements and prizes for participants [4][6]. - The event will feature experts from Keysight, including application engineers and product managers, who will discuss the application of artificial neural networks and the ML Optimizer in device modeling [8][11].